CareProof

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for CareProof

Inspiration

Care organizations create valuable frontline records every day, but turning those notes into a report for a client, partner, or supervisor is slow and risky. Personal identifiers can leak into shared documents, broad conclusions can outrun the evidence, and polished summaries often lose the connection to their source. CareProof was inspired by a simple question: can reporting become faster without asking teams to trade away privacy, provenance, or professional judgment?

What it does

CareProof converts pasted notes or CSV/TXT records into a privacy-safe, evidence-linked report. It removes phone numbers, participant details, and internal staff identifiers; flags unsupported or overly confident conclusions; and gives each report statement a source reference. Reviewers can compare the original observation with its share-safe version, switch between English and Japanese, copy the report, export a sanitized CSV, or print to PDF. All note processing runs locally in the browser.

How we built it

The product was built with OpenAI Codex and GPT-5.6 as the primary development environment. Codex helped translate the reporting workflow into product requirements, implement the TypeScript analysis pipeline and bilingual React interface, create privacy and claim-safety checks, design the evidence map, write tests, and prepare the production build. The app uses Next.js, React, TypeScript, Vinext, and Cloudflare-compatible deployment tooling. Its current analysis is deliberately deterministic and local-first so judges can inspect every transformation without sending sensitive notes to an external service.

Challenges we ran into

The hardest product challenge was balancing useful summarization with restraint. A care report must be readable, but it must not turn an observation into a diagnosis or an unsupported outcome claim. We therefore designed the output around bounded language, explicit evidence references, and human review instead of pretending the tool can make professional decisions. We also had to make privacy transformations visible enough to audit without exposing the redacted value again.

Accomplishments that we're proud of

CareProof is a working bilingual prototype with an end-to-end review flow. A user can load fictional care notes, see a report immediately, trace every statement back to a source line, review privacy redactions and risky language, and export a safe dataset. The interface communicates when review is still required and clearly states that the tool is decision support rather than medical advice.

What we learned

Trustworthy reporting is not just a text-generation problem. It is a provenance, privacy, and interaction-design problem. Users need to understand what changed, why it changed, and which statements remain uncertain. Codex was especially useful for moving between those layers—product logic, safety rules, interface behavior, testing, and deployment—inside one continuous build session.

What's next for CareProof

Next, we would add configurable organization policies, richer CSV column mapping, role-based review, encrypted saved projects, signed export history, and optional model-assisted drafting with strict citation constraints. We would also validate the workflow with care managers and privacy officers before any use with real personal data.

Built With

• openai
• typescript